The challenges of small targets and complex backgrounds in traffic sign detection cannot be ef-fectively addressed by traditional methods.To this end,this paper proposes a traffic sign detection algo-rithm based on feature enhancement and attention focusing.Firstly,a dual-backbone network is designed to fuse and enhance the shallow feature information and the deep semantic information,so as to alleviate the problem of information loss and insufficient fusion of the neck network feature information due to multiple down sampling.By adding a small target detection layer,it is better adapted to the detection of small target traffic signs in natural scenes.Secondly,the CA3 attention module was designed to complement the chan-nel features and spatial position information into the feature map to reduce the noise interference of complex backgrounds.Finally,WIoUv3 Loss was used to dynamically allocate the gradient gain reasonably and re-duce the harmful gradient generated by low-quality samples in the data set in time.Experimental results show that the proposed model achieves an average accuracy(mAP)improvement of 4.30％and 1.10％compared with YOLOv5 on the CCTSDB 2021 and GTSDB datasets,respectively.Compared with other ma-instream models,this model shows better detection of traffic signs in complex road scenarios.
维普
万方数据